LaunchWe're giving away FREE LIFETIME subscriptions for the first 200 users!--:--:--Add to Chrome →
FindcheapFindcheap
Add to Chrome

Introducing Findcheap

PriceBench Performance

0%25%50%75%100%0s25s50s75s100smedian search latencysavings captureFindcheap15.6s · 92.8%GPT-6 Astra87.5s · 43.9%Perplexity74.7s · 34.1%Gemini 3.1 Pro80s · 10.7%
PriceBench results, equivalent-product track, human-verified, N=100 products. Savings capture plotted against median search latency for the interactive agents.
Just want to try it? Click here.

Today, we're excited to announce that Findcheap has scored more than 2x higher than the leading frontier model on product search, which is widely-believed to be the single most important web search task.

What is Findcheap?

Findcheap is an AI agent that searches the internet to find the cheapest place to buy stuff.

What makes it unique is that it thinks like a human does. It's able to find cheaper options for 94% of products on the internet, and saves users 60.4% on average.

How do I use it?

Just install the Chrome extension and shop like you normally do. Findcheap will automatically detect when you're on a product page, and proactively offer to help.

We're releasing two products today: a consumer product (Chrome extension) and an infra product (API). Exciting things coming soon, but that's a post for a different day!

How is Findcheap able to beat the big AI labs?

Frontier models try to be the best at everything. Findcheap tries to be the best at a narrow subfield of agentic AI: multimodal search.

Task: find lower-priced equivalent products

0%25%50%75%100%92.8%43.9%34.1%20.2%10.7%

Task: find exact brand/model for the lowest available price

0%25%50%75%100%79.6%52.5%20.7%24.9%10.7%

Latency

(lower is better)
0s60s120s180s240s15.6s87.5s74.7s232s80s

Cost

(lower is better)
$0.00$0.50$1.00$1.50$0.03$1.32$0.38$1.41$0.08
PriceBench N=100 human-verified results: savings capture on both tracks, median latency, and cost per search for Findcheap, GPT-6 Astra, Perplexity, Claude Opus 5, and Gemini 3.1 Pro.

Note: the pricing shown above is for the Findcheap API. Findcheap is completely free through the Chrome extension.

How was model performance measured?

Because this technology is still early, there's no widely-used benchmark for evaluating agents’ performance on this task. That's why PriceBench was created. This benchmark is the first of its kind in a number of ways:

  • It doesn't have a fixed set of input tasks, which makes it impossible to game the benchmark by overfitting on specific tasks. Instead, it autonomously sources new input tasks at runtime.
  • It's designed to run with almost no human intervention. A panel of 3 judges composed of GPT-5.6 Sol, Claude Opus 5, and Gemini 3.5 Flash autonomously reviews results, to avoid human biases.
  • It sources tasks based on real-world data. For a benchmark to accurately reflect a real user’s experience, it must include a mix of large and small retailers, and different product categories. For example, Amazon accounts for about 42% of e-commerce sales in the US, so 42 of 100 input tasks are sourced from Amazon.

In the interest of encouraging innovation in this space, the benchmark is open source and includes a skill to run it in one prompt! Check it out on GitHub.